Binding free energy calculation and scoring in small-molecule docking

CAVASOTTO, CLAUDIO N.

Physico-Chemical and Computational Appraoches to Drug Discovery

Royal Society of Chemistry

Año: 2012; p. 195 - 222

The correct functioning of the biochemical machinery within a cell depends upon the non-covalent molecular association in processes such as enzyme catalysis, molecular transport and signal transduction. Until recently, High Throughput Screening (HTS) has been the dominant technique within pharmaceutical companies to identify and optimize new drug lead compounds.  HTS involves combinatorial chemistry and the experimental screening of large chemical libraries against a relevant therapeutic target. It is thus an expensive and time-consuming process, in spite of recent progresses to improve its efficiency. The use of 3D structures of protein-ligand complexes have been used since some time ago to guide drug lead optimization aiming to improve potency or selectivity, thus giving rise to a more rational appraoch.  The natural contituation of this process was the development of in silico methods, both as a means to predict protein-ligand interactions, and for the computational screen of chemical libraries through protein-ligand docking, a process in which the ligand is positioned within the binding site, and its binding energy estimated.  This computational mimic of HTS provided a more rational, more economic, and faster alternative to the traditional HTS.  It can be said that currently structure-based drug lead discovery and design is a key first step in the lengthy, expensive and unpredicable process of developing new drugs.   This development of in silico tools to study protein-ligand interactions, and to dock chemical libraries began in the late 1970s and has been in constant progress ever since. The main advantages of docking-based methods compared to ligand-based methods are the structural novelty of the hits discovered ?not based on pre-existing, known ligands-, and the possibility to model the binding mode of potential ligands within the binding site.  Although high-throughput docking (HTD) is plagued with both false-positives and false-negatives, the low set-up cost, high computational speed, potential structural novelty of discovered ligands, and the possibility to incorporate at any stage of the process filters accounting for drug-likeness offset its limitations.  In computational docking studies, three main ideal purposes could be identified: i) The correct characterization of the binding pose (or the thermodynamical equilibrium ensemble) of a known ligand within the binding site (in principle, this involves the degrees of freedom (DOF) of both the ligand and the receptor); ii) the accurate calculation of the ligand binding free energy, or the relative free energies of a series of molecules; iii) the prediction of the pose and estimation of the binding free energy of large virtual chemical libraries in a high-throughput fashion (HTD). The prediction of the dominant binding pose, or the equilibrium ensemble that characterizes the ligand-protein complex can be determined using only the potential energy surface of the complex, for example, from molecular dynamics (MD) or Monte Carlo (MC) simulations.  However, in order to use docking in a high-throughput fashion, several approximations should be introduced to reduce the number of DOF to be sampled.  Thus, in HTD, the receptor is usually considered rigid (or very few protein DOF considered), solvent is represented in a continuum fashion, and hard DOF (bond lenghts and planar angles of the ligand) are frozen.  In spite of these approximations, ligand posing through docking has shown a reasonable degree of success, and is not the main limitation step in the application of HTD to drug discovery.  It should be remarked that other factors may also condition docking accuracy, such as the quality of the receptor structure used, the type of receptor used -bound (holo) or unbound-, the inclusion of crystallographic water molecules, and the correct assignment of the protonation and tautomerization states of the ligand and binding site residues.  These latter factors are not approximations themselves, but reflect uncertainties in the description of the molecular system. This chapter will open with a review of the theory of binding free energy and its computational calculation. The description of the different types of scoring functions in HTD will follow, giving consideration to comparison studies, assessing their limitations and the facts that hinder their improvement.  The post-screening process will be described next, with special emphasis in the use of advanced simulation methods to re-score top-ranking hits. The conclusions and suggested future directions will close this chapter.